simplegrade 0.1.3

A lightweight PyTorch-like autograd library built from scratch

SimpleGrad

SimpleGrad is a lightweight Python library for automatic differentiation and neural networks.
It is inspired by PyTorch and Tinygrad, offering a balance between simplicity, flexibility, and educational value.

---

Features

• Tensor operations: +, -, *, /, @ (matmul), pow, exp, log, ln, sin, cos, tanh, relu, sigmoid.
• Automatic differentiation: .backward() computes gradients for all operations in the computation graph.
• Neural networks: Build networks using Linear layers and activation functions.
• Loss functions: MSE, MAE, etc. (can be extended).
• Gradient descent: Manual update or custom optimizers can be implemented.
• Minimal and educational: Perfect for learning how autograd works under the hood.

---

from simplegrade import Tensor

# --------------------------
# Tensor operations
# --------------------------
a = Tensor([[1, 2, 3]], requires_grad=True)
b = Tensor([[2, 1, 0]], requires_grad=True)
c = a + b
d = a * b
e = d.sum()

print("c:", c)
print("d:", d)
print("e:", e)

# Backward pass
e.backward()
print("grad a:", a.grad)
print("grad b:", b.grad)

Installation

Clone the repository and install in editable mode:

pip install simplegrade